Mapping Analysis to Predict the Associated EQ-5D-3L Utility Values from the Oxford Hip Score

Purpose: The aim was to assess mapping models to predict the EQ-5D-3L utility from the Oxford hip score (OHS) and validate these before and after total hip arthroplasty (THA). Methods: Separate retrospective cohorts of patients who had undergone THA in different centres were used to create (n = 3981) and validate (n = 7168) the prediction models. Patient demographics, body mass index (BMI), OHS and EQ-5D were collected preoperatively and one year postoperatively. Predictive models were formulated using generalised linear regression. Results: In the modelling cohort, mean OHS improved from 20.8 preoperatively to 39.6 postoperatively; the mean EQ-5D improved from 0.391 preoperatively to 0.788 postoperatively. Significant (p < 0.001) correlations were observed between OHS and EQ-5D preoperatively (r = 0.690) and postoperatively (r = 0.777) and for change (r = 0.611) and combined scores (r = 0.812). Four models predicting EQ-5D from OHS explained 38.5% to 65.9% of the variance. Age and BMI did not improve model fit. Predicted mean EQ-5D utilities differed statistically from observed values for most models, but all differences were below the minimal clinical important difference (MCID). Including individual OHS items marginally improved performance (<2%). External validation showed good agreement at the cohort level but lower accuracy for individual-level prediction (27.7% for change to 65.0% for postoperative scores to be within the MCID). Conclusions: The EQ-5D-3L utility index, on a cohort level, may be reliably estimated from the OHS. Predicted individual patient utility error beyond the MCID varied from one-third to more than two-thirds depending on the data assessed and so should not be employed for this purpose on a patient level.